1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2007, 2009, 2010, 2011 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19 #include "data/sys-file-writer.h"
20 #include "data/sys-file-private.h"
29 #include "data/attributes.h"
30 #include "data/case.h"
31 #include "data/casewriter-provider.h"
32 #include "data/casewriter.h"
33 #include "data/dictionary.h"
34 #include "data/file-handle-def.h"
35 #include "data/file-name.h"
36 #include "data/format.h"
37 #include "data/make-file.h"
38 #include "data/missing-values.h"
39 #include "data/mrset.h"
40 #include "data/settings.h"
41 #include "data/short-names.h"
42 #include "data/value-labels.h"
43 #include "data/variable.h"
44 #include "libpspp/float-format.h"
45 #include "libpspp/i18n.h"
46 #include "libpspp/integer-format.h"
47 #include "libpspp/message.h"
48 #include "libpspp/misc.h"
49 #include "libpspp/str.h"
50 #include "libpspp/string-array.h"
51 #include "libpspp/version.h"
53 #include "gl/xmemdup0.h"
54 #include "gl/minmax.h"
55 #include "gl/unlocked-io.h"
56 #include "gl/xalloc.h"
59 #define _(msgid) gettext (msgid)
60 #define N_(msgid) (msgid)
62 /* Compression bias used by PSPP. Values between (1 -
63 COMPRESSION_BIAS) and (251 - COMPRESSION_BIAS) inclusive can be
65 #define COMPRESSION_BIAS 100
67 /* System file writer. */
70 struct file_handle *fh; /* File handle. */
71 struct fh_lock *lock; /* Mutual exclusion for file. */
72 FILE *file; /* File stream. */
73 struct replace_file *rf; /* Ticket for replacing output file. */
75 bool compress; /* 1=compressed, 0=not compressed. */
76 casenumber case_cnt; /* Number of cases written so far. */
78 /* Compression buffering.
80 Compressed data is output as groups of 8 1-byte opcodes
81 followed by up to 8 (depending on the opcodes) 8-byte data
82 items. Data items and opcodes arrive at the same time but
83 must be reordered for writing to disk, thus a small amount
85 uint8_t opcodes[8]; /* Buffered opcodes. */
86 int opcode_cnt; /* Number of buffered opcodes. */
87 uint8_t data[8][8]; /* Buffered data. */
88 int data_cnt; /* Number of buffered data items. */
91 struct sfm_var *sfm_vars; /* Variables. */
92 size_t sfm_var_cnt; /* Number of variables. */
93 size_t segment_cnt; /* Number of variables including extra segments
94 for long string variables. */
97 static const struct casewriter_class sys_file_casewriter_class;
99 static void write_header (struct sfm_writer *, const struct dictionary *);
100 static void write_variable (struct sfm_writer *, const struct variable *);
101 static void write_value_labels (struct sfm_writer *, struct variable *,
103 static void write_integer_info_record (struct sfm_writer *);
104 static void write_float_info_record (struct sfm_writer *);
106 static void write_longvar_table (struct sfm_writer *w,
107 const struct dictionary *dict);
109 static void write_encoding_record (struct sfm_writer *w,
110 const struct dictionary *);
112 static void write_vls_length_table (struct sfm_writer *w,
113 const struct dictionary *dict);
115 static void write_long_string_value_labels (struct sfm_writer *,
116 const struct dictionary *);
118 static void write_mrsets (struct sfm_writer *, const struct dictionary *,
121 static void write_variable_display_parameters (struct sfm_writer *w,
122 const struct dictionary *dict);
124 static void write_documents (struct sfm_writer *, const struct dictionary *);
126 static void write_data_file_attributes (struct sfm_writer *,
127 const struct dictionary *);
128 static void write_variable_attributes (struct sfm_writer *,
129 const struct dictionary *);
131 static void write_int (struct sfm_writer *, int32_t);
132 static inline void convert_double_to_output_format (double, uint8_t[8]);
133 static void write_float (struct sfm_writer *, double);
134 static void write_string (struct sfm_writer *, const char *, size_t);
135 static void write_utf8_string (struct sfm_writer *, const char *encoding,
136 const char *string, size_t width);
137 static void write_utf8_record (struct sfm_writer *, const char *encoding,
138 const struct string *content, int subtype);
139 static void write_string_record (struct sfm_writer *,
140 const struct substring content, int subtype);
141 static void write_bytes (struct sfm_writer *, const void *, size_t);
142 static void write_zeros (struct sfm_writer *, size_t);
143 static void write_spaces (struct sfm_writer *, size_t);
144 static void write_value (struct sfm_writer *, const union value *, int width);
146 static void write_case_uncompressed (struct sfm_writer *,
147 const struct ccase *);
148 static void write_case_compressed (struct sfm_writer *, const struct ccase *);
149 static void flush_compressed (struct sfm_writer *);
150 static void put_cmp_opcode (struct sfm_writer *, uint8_t);
151 static void put_cmp_number (struct sfm_writer *, double);
152 static void put_cmp_string (struct sfm_writer *, const void *, size_t);
154 static bool write_error (const struct sfm_writer *);
155 static bool close_writer (struct sfm_writer *);
157 /* Returns default options for writing a system file. */
158 struct sfm_write_options
159 sfm_writer_default_options (void)
161 struct sfm_write_options opts;
162 opts.create_writeable = true;
163 opts.compress = settings_get_scompression ();
168 /* Opens the system file designated by file handle FH for writing
169 cases from dictionary D according to the given OPTS.
171 No reference to D is retained, so it may be modified or
172 destroyed at will after this function returns. D is not
173 modified by this function, except to assign short names. */
175 sfm_open_writer (struct file_handle *fh, struct dictionary *d,
176 struct sfm_write_options opts)
178 struct sfm_writer *w;
184 if (opts.version != 2 && opts.version != 3)
186 msg (ME, _("Unknown system file version %d. Treating as version %d."),
191 /* Create and initialize writer. */
192 w = xmalloc (sizeof *w);
198 w->compress = opts.compress;
201 w->opcode_cnt = w->data_cnt = 0;
203 /* Figure out how to map in-memory case data to on-disk case
204 data. Also count the number of segments. Very long strings
205 occupy multiple segments, otherwise each variable only takes
207 w->segment_cnt = sfm_dictionary_to_sfm_vars (d, &w->sfm_vars,
210 /* Open file handle as an exclusive writer. */
211 /* TRANSLATORS: this fragment will be interpolated into
212 messages in fh_lock() that identify types of files. */
213 w->lock = fh_lock (fh, FH_REF_FILE, N_("system file"), FH_ACC_WRITE, true);
217 /* Create the file on disk. */
219 if (opts.create_writeable)
221 w->rf = replace_file_start (fh_get_file_name (fh), "wb", mode,
225 msg (ME, _("Error opening `%s' for writing as a system file: %s."),
226 fh_get_file_name (fh), strerror (errno));
230 /* Write the file header. */
233 /* Write basic variable info. */
234 short_names_assign (d);
235 for (i = 0; i < dict_get_var_cnt (d); i++)
236 write_variable (w, dict_get_var (d, i));
238 /* Write out value labels. */
240 for (i = 0; i < dict_get_var_cnt (d); i++)
242 struct variable *v = dict_get_var (d, i);
244 write_value_labels (w, v, idx);
245 idx += sfm_width_to_octs (var_get_width (v));
248 if (dict_get_document_line_cnt (d) > 0)
249 write_documents (w, d);
251 write_integer_info_record (w);
252 write_float_info_record (w);
254 write_mrsets (w, d, true);
256 write_variable_display_parameters (w, d);
258 if (opts.version >= 3)
259 write_longvar_table (w, d);
261 write_vls_length_table (w, d);
263 write_long_string_value_labels (w, d);
265 if (attrset_count (dict_get_attributes (d)))
266 write_data_file_attributes (w, d);
267 write_variable_attributes (w, d);
269 write_mrsets (w, d, false);
271 write_encoding_record (w, d);
273 /* Write end-of-headers record. */
280 return casewriter_create (dict_get_proto (d), &sys_file_casewriter_class, w);
287 /* Returns value of X truncated to two least-significant digits. */
298 /* Calculates the offset of data for TARGET_VAR from the
299 beginning of each case's data for dictionary D. The return
300 value is in "octs" (8-byte units). */
302 calc_oct_idx (const struct dictionary *d, struct variable *target_var)
308 for (i = 0; i < dict_get_var_cnt (d); i++)
310 struct variable *var = dict_get_var (d, i);
311 if (var == target_var)
313 oct_idx += sfm_width_to_octs (var_get_width (var));
318 /* Write the sysfile_header header to system file W. */
320 write_header (struct sfm_writer *w, const struct dictionary *d)
323 char creation_date[10];
324 char creation_time[9];
325 const char *file_label;
326 struct variable *weight;
330 /* Record-type code. */
331 write_string (w, "$FL2", 4);
333 /* Product identification. */
334 snprintf (prod_name, sizeof prod_name, "@(#) SPSS DATA FILE %s - %s",
335 version, host_system);
336 write_string (w, prod_name, 60);
341 /* Number of `union value's per case. */
342 write_int (w, calc_oct_idx (d, NULL));
345 write_int (w, w->compress);
347 /* Weight variable. */
348 weight = dict_get_weight (d);
349 write_int (w, weight != NULL ? calc_oct_idx (d, weight) + 1 : 0);
351 /* Number of cases. We don't know this in advance, so we write
352 -1 to indicate an unknown number of cases. Later we can
353 come back and overwrite it with the true value. */
356 /* Compression bias. */
357 write_float (w, COMPRESSION_BIAS);
359 /* Creation date and time. */
360 if (time (&t) == (time_t) -1)
362 strcpy (creation_date, "01 Jan 70");
363 strcpy (creation_time, "00:00:00");
367 static const char *const month_name[12] =
369 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
370 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec",
372 struct tm *tmp = localtime (&t);
373 int day = rerange (tmp->tm_mday);
374 int mon = rerange (tmp->tm_mon + 1);
375 int year = rerange (tmp->tm_year);
376 int hour = rerange (tmp->tm_hour + 1);
377 int min = rerange (tmp->tm_min + 1);
378 int sec = rerange (tmp->tm_sec + 1);
380 snprintf (creation_date, sizeof creation_date,
381 "%02d %s %02d", day, month_name[mon - 1], year);
382 snprintf (creation_time, sizeof creation_time,
383 "%02d:%02d:%02d", hour - 1, min - 1, sec - 1);
385 write_string (w, creation_date, 9);
386 write_string (w, creation_time, 8);
389 file_label = dict_get_label (d);
390 if (file_label == NULL)
392 write_utf8_string (w, dict_get_encoding (d), file_label, 64);
398 /* Write format spec FMT to W, after adjusting it to be
399 compatible with the given WIDTH. */
401 write_format (struct sfm_writer *w, struct fmt_spec fmt, int width)
403 assert (fmt_check_output (&fmt));
404 assert (sfm_width_to_segments (width) == 1);
407 fmt_resize (&fmt, width);
408 write_int (w, (fmt_to_io (fmt.type) << 16) | (fmt.w << 8) | fmt.d);
411 /* Write a string continuation variable record for each 8-byte
412 section beyond the initial 8 bytes, for a variable of the
415 write_variable_continuation_records (struct sfm_writer *w, int width)
419 assert (sfm_width_to_segments (width) == 1);
420 for (position = 8; position < width; position += 8)
422 write_int (w, 2); /* Record type. */
423 write_int (w, -1); /* Width. */
424 write_int (w, 0); /* No variable label. */
425 write_int (w, 0); /* No missing values. */
426 write_int (w, 0); /* Print format. */
427 write_int (w, 0); /* Write format. */
428 write_zeros (w, 8); /* Name. */
432 /* Write the variable record(s) for variable V to system file
435 write_variable (struct sfm_writer *w, const struct variable *v)
437 int width = var_get_width (v);
438 int segment_cnt = sfm_width_to_segments (width);
439 int seg0_width = sfm_segment_alloc_width (width, 0);
440 const char *encoding = var_get_encoding (v);
441 struct missing_values mv;
448 write_int (w, seg0_width);
450 /* Variable has a variable label? */
451 write_int (w, var_has_label (v));
453 /* Number of missing values. If there is a range, then the
454 range counts as 2 missing values and causes the number to be
456 mv_copy (&mv, var_get_missing_values (v));
457 if (mv_get_width (&mv) > 8)
459 if (mv_has_range (&mv))
460 write_int (w, -2 - mv_n_values (&mv));
462 write_int (w, mv_n_values (&mv));
464 /* Print and write formats. */
465 write_format (w, *var_get_print_format (v), seg0_width);
466 write_format (w, *var_get_write_format (v), seg0_width);
469 The full name is in a translation table written
471 write_utf8_string (w, encoding, var_get_short_name (v, 0), 8);
474 if (var_has_label (v))
476 char *label = recode_string (encoding, UTF8, var_get_label (v), -1);
477 size_t label_len = MIN (strlen (label), 255);
478 size_t padded_len = ROUND_UP (label_len, 4);
479 write_int (w, label_len);
480 write_string (w, label, padded_len);
484 /* Write the missing values, if any, range first. */
485 if (mv_has_range (&mv))
488 mv_get_range (&mv, &x, &y);
492 for (i = 0; i < mv_n_values (&mv); i++)
493 write_value (w, mv_get_value (&mv, i), mv_get_width (&mv));
495 write_variable_continuation_records (w, seg0_width);
497 /* Write additional segments for very long string variables. */
498 for (i = 1; i < segment_cnt; i++)
500 int seg_width = sfm_segment_alloc_width (width, i);
501 struct fmt_spec fmt = fmt_for_output (FMT_A, MAX (seg_width, 1), 0);
503 write_int (w, 2); /* Variable record. */
504 write_int (w, seg_width); /* Width. */
505 write_int (w, 0); /* No variable label. */
506 write_int (w, 0); /* No missing values. */
507 write_format (w, fmt, seg_width); /* Print format. */
508 write_format (w, fmt, seg_width); /* Write format. */
509 write_utf8_string (w, encoding, var_get_short_name (v, i), 8);
511 write_variable_continuation_records (w, seg_width);
517 /* Writes the value labels for variable V having system file
518 variable index IDX to system file W.
520 Value labels for long string variables are written separately,
521 by write_long_string_value_labels. */
523 write_value_labels (struct sfm_writer *w, struct variable *v, int idx)
525 const struct val_labs *val_labs;
526 const struct val_lab **labels;
530 val_labs = var_get_value_labels (v);
531 n_labels = val_labs_count (val_labs);
532 if (n_labels == 0 || var_get_width (v) > 8)
535 /* Value label record. */
536 write_int (w, 3); /* Record type. */
537 write_int (w, val_labs_count (val_labs));
538 labels = val_labs_sorted (val_labs);
539 for (i = 0; i < n_labels; i++)
541 const struct val_lab *vl = labels[i];
542 char *label = recode_string (var_get_encoding (v), UTF8,
543 val_lab_get_escaped_label (vl), -1);
544 uint8_t len = MIN (strlen (label), 255);
546 write_value (w, val_lab_get_value (vl), var_get_width (v));
547 write_bytes (w, &len, 1);
548 write_bytes (w, label, len);
549 write_zeros (w, REM_RND_UP (len + 1, 8));
554 /* Value label variable record. */
555 write_int (w, 4); /* Record type. */
556 write_int (w, 1); /* Number of variables. */
557 write_int (w, idx + 1); /* Variable's dictionary index. */
560 /* Writes record type 6, document record. */
562 write_documents (struct sfm_writer *w, const struct dictionary *d)
564 const struct string_array *docs = dict_get_documents (d);
565 const char *enc = dict_get_encoding (d);
568 write_int (w, 6); /* Record type. */
569 write_int (w, docs->n);
570 for (i = 0; i < docs->n; i++)
572 char *s = recode_string (enc, "UTF-8", docs->strings[i], -1);
573 size_t s_len = strlen (s);
574 size_t write_len = MIN (s_len, DOC_LINE_LENGTH);
576 write_bytes (w, s, write_len);
577 write_spaces (w, DOC_LINE_LENGTH - write_len);
583 put_attrset (struct string *string, const struct attrset *attrs)
585 const struct attribute *attr;
586 struct attrset_iterator i;
588 for (attr = attrset_first (attrs, &i); attr != NULL;
589 attr = attrset_next (attrs, &i))
591 size_t n_values = attribute_get_n_values (attr);
594 ds_put_cstr (string, attribute_get_name (attr));
595 ds_put_byte (string, '(');
596 for (j = 0; j < n_values; j++)
597 ds_put_format (string, "'%s'\n", attribute_get_value (attr, j));
598 ds_put_byte (string, ')');
603 write_data_file_attributes (struct sfm_writer *w,
604 const struct dictionary *d)
606 struct string s = DS_EMPTY_INITIALIZER;
607 put_attrset (&s, dict_get_attributes (d));
608 write_utf8_record (w, dict_get_encoding (d), &s, 17);
613 write_variable_attributes (struct sfm_writer *w, const struct dictionary *d)
615 struct string s = DS_EMPTY_INITIALIZER;
616 size_t n_vars = dict_get_var_cnt (d);
617 size_t n_attrsets = 0;
620 for (i = 0; i < n_vars; i++)
622 struct variable *v = dict_get_var (d, i);
623 struct attrset *attrs = var_get_attributes (v);
624 if (attrset_count (attrs))
627 ds_put_byte (&s, '/');
628 ds_put_format (&s, "%s:", var_get_short_name (v, 0));
629 put_attrset (&s, attrs);
633 write_utf8_record (w, dict_get_encoding (d), &s, 18);
637 /* Write multiple response sets. If PRE_V14 is true, writes sets supported by
638 SPSS before release 14, otherwise writes sets supported only by later
641 write_mrsets (struct sfm_writer *w, const struct dictionary *dict,
644 const char *encoding = dict_get_encoding (dict);
645 struct string s = DS_EMPTY_INITIALIZER;
649 n_mrsets = dict_get_n_mrsets (dict);
653 for (i = 0; i < n_mrsets; i++)
655 const struct mrset *mrset = dict_get_mrset (dict, i);
659 if ((mrset->type != MRSET_MD || mrset->cat_source != MRSET_COUNTEDVALUES)
663 name = recode_string (encoding, "UTF-8", mrset->name, -1);
664 ds_put_format (&s, "%s=", name);
667 if (mrset->type == MRSET_MD)
671 if (mrset->cat_source == MRSET_COUNTEDVALUES)
672 ds_put_format (&s, "E %d ", mrset->label_from_var_label ? 11 : 1);
674 ds_put_byte (&s, 'D');
676 if (mrset->width == 0)
677 counted = xasprintf ("%.0f", mrset->counted.f);
679 counted = xmemdup0 (value_str (&mrset->counted, mrset->width),
681 ds_put_format (&s, "%zu %s", strlen (counted), counted);
685 ds_put_byte (&s, 'C');
686 ds_put_byte (&s, ' ');
688 if (mrset->label && !mrset->label_from_var_label)
690 char *label = recode_string (encoding, "UTF-8", mrset->label, -1);
691 ds_put_format (&s, "%zu %s", strlen (label), label);
695 ds_put_cstr (&s, "0 ");
697 for (j = 0; j < mrset->n_vars; j++)
699 const char *short_name_utf8 = var_get_short_name (mrset->vars[j], 0);
700 char *short_name = recode_string (encoding, "UTF-8",
701 short_name_utf8, -1);
702 ds_put_format (&s, " %s", short_name);
705 ds_put_byte (&s, '\n');
708 write_string_record (w, ds_ss (&s), pre_v14 ? 7 : 19);
712 /* Write the alignment, width and scale values. */
714 write_variable_display_parameters (struct sfm_writer *w,
715 const struct dictionary *dict)
719 write_int (w, 7); /* Record type. */
720 write_int (w, 11); /* Record subtype. */
721 write_int (w, 4); /* Data item (int32) size. */
722 write_int (w, w->segment_cnt * 3); /* Number of data items. */
724 for (i = 0; i < dict_get_var_cnt (dict); ++i)
726 struct variable *v = dict_get_var (dict, i);
727 int width = var_get_width (v);
728 int segment_cnt = sfm_width_to_segments (width);
729 int measure = (var_get_measure (v) == MEASURE_NOMINAL ? 1
730 : var_get_measure (v) == MEASURE_ORDINAL ? 2
732 int alignment = (var_get_alignment (v) == ALIGN_LEFT ? 0
733 : var_get_alignment (v) == ALIGN_RIGHT ? 1
737 for (i = 0; i < segment_cnt; i++)
739 int width_left = width - sfm_segment_effective_offset (width, i);
740 write_int (w, measure);
741 write_int (w, (i == 0 ? var_get_display_width (v)
742 : var_default_display_width (width_left)));
743 write_int (w, alignment);
748 /* Writes the table of lengths for very long string variables. */
750 write_vls_length_table (struct sfm_writer *w,
751 const struct dictionary *dict)
756 ds_init_empty (&map);
757 for (i = 0; i < dict_get_var_cnt (dict); ++i)
759 const struct variable *v = dict_get_var (dict, i);
760 if (sfm_width_to_segments (var_get_width (v)) > 1)
761 ds_put_format (&map, "%s=%05d%c\t",
762 var_get_short_name (v, 0), var_get_width (v), 0);
764 if (!ds_is_empty (&map))
765 write_utf8_record (w, dict_get_encoding (dict), &map, 14);
771 write_long_string_value_labels (struct sfm_writer *w,
772 const struct dictionary *dict)
774 size_t n_vars = dict_get_var_cnt (dict);
778 /* Figure out the size in advance. */
780 for (i = 0; i < n_vars; i++)
782 struct variable *var = dict_get_var (dict, i);
783 const struct val_labs *val_labs = var_get_value_labels (var);
784 const char *encoding = var_get_encoding (var);
785 int width = var_get_width (var);
786 const struct val_lab *val_lab;
788 if (val_labs_count (val_labs) == 0 || width < 9)
792 size += recode_string_len (encoding, "UTF-8", var_get_name (var), -1);
793 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
794 val_lab = val_labs_next (val_labs, val_lab))
797 size += recode_string_len (encoding, "UTF-8",
798 val_lab_get_escaped_label (val_lab), -1);
804 write_int (w, 7); /* Record type. */
805 write_int (w, 21); /* Record subtype */
806 write_int (w, 1); /* Data item (byte) size. */
807 write_int (w, size); /* Number of data items. */
809 start = ftello (w->file);
810 for (i = 0; i < n_vars; i++)
812 struct variable *var = dict_get_var (dict, i);
813 const struct val_labs *val_labs = var_get_value_labels (var);
814 const char *encoding = var_get_encoding (var);
815 int width = var_get_width (var);
816 const struct val_lab *val_lab;
819 if (val_labs_count (val_labs) == 0 || width < 9)
822 var_name = recode_string (encoding, "UTF-8", var_get_name (var), -1);
823 write_int (w, strlen (var_name));
824 write_bytes (w, var_name, strlen (var_name));
827 write_int (w, width);
828 write_int (w, val_labs_count (val_labs));
829 for (val_lab = val_labs_first (val_labs); val_lab != NULL;
830 val_lab = val_labs_next (val_labs, val_lab))
835 write_int (w, width);
836 write_bytes (w, value_str (val_lab_get_value (val_lab), width),
839 label = recode_string (var_get_encoding (var), "UTF-8",
840 val_lab_get_escaped_label (val_lab), -1);
841 len = strlen (label);
843 write_bytes (w, label, len);
847 assert (ftello (w->file) == start + size);
851 write_encoding_record (struct sfm_writer *w,
852 const struct dictionary *d)
854 const char *enc = dict_get_encoding (d);
859 write_int (w, 7); /* Record type. */
860 write_int (w, 20); /* Record subtype. */
861 write_int (w, 1); /* Data item (char) size. */
863 /* IANA says "...character set names may be up to 40 characters taken from
864 the printable characters of US-ASCII," so character set names don't need
866 write_int (w, strlen (enc)); /* Number of data items. */
867 write_string (w, enc, strlen (enc));
871 /* Writes the long variable name table. */
873 write_longvar_table (struct sfm_writer *w, const struct dictionary *dict)
878 ds_init_empty (&map);
879 for (i = 0; i < dict_get_var_cnt (dict); i++)
881 struct variable *v = dict_get_var (dict, i);
883 ds_put_byte (&map, '\t');
884 ds_put_format (&map, "%s=%s",
885 var_get_short_name (v, 0), var_get_name (v));
887 write_utf8_record (w, dict_get_encoding (dict), &map, 13);
891 /* Write integer information record. */
893 write_integer_info_record (struct sfm_writer *w)
895 int version_component[3];
898 /* Parse the version string. */
899 memset (version_component, 0, sizeof version_component);
900 sscanf (bare_version, "%d.%d.%d",
901 &version_component[0], &version_component[1], &version_component[2]);
903 /* Figure out the floating-point format. */
904 if (FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_LE
905 || FLOAT_NATIVE_64_BIT == FLOAT_IEEE_DOUBLE_BE)
907 else if (FLOAT_NATIVE_64_BIT == FLOAT_Z_LONG)
909 else if (FLOAT_NATIVE_64_BIT == FLOAT_VAX_D)
915 write_int (w, 7); /* Record type. */
916 write_int (w, 3); /* Record subtype. */
917 write_int (w, 4); /* Data item (int32) size. */
918 write_int (w, 8); /* Number of data items. */
919 write_int (w, version_component[0]);
920 write_int (w, version_component[1]);
921 write_int (w, version_component[2]);
922 write_int (w, -1); /* Machine code. */
923 write_int (w, float_format);
924 write_int (w, 1); /* Compression code. */
925 write_int (w, INTEGER_NATIVE == INTEGER_MSB_FIRST ? 1 : 2);
926 write_int (w, 2); /* 7-bit ASCII. */
929 /* Write floating-point information record. */
931 write_float_info_record (struct sfm_writer *w)
933 write_int (w, 7); /* Record type. */
934 write_int (w, 4); /* Record subtype. */
935 write_int (w, 8); /* Data item (flt64) size. */
936 write_int (w, 3); /* Number of data items. */
937 write_float (w, SYSMIS); /* System-missing value. */
938 write_float (w, HIGHEST); /* Value used for HIGHEST in missing values. */
939 write_float (w, LOWEST); /* Value used for LOWEST in missing values. */
942 /* Writes case C to system file W. */
944 sys_file_casewriter_write (struct casewriter *writer, void *w_,
947 struct sfm_writer *w = w_;
949 if (ferror (w->file))
951 casewriter_force_error (writer);
959 write_case_uncompressed (w, c);
961 write_case_compressed (w, c);
966 /* Destroys system file writer W. */
968 sys_file_casewriter_destroy (struct casewriter *writer, void *w_)
970 struct sfm_writer *w = w_;
971 if (!close_writer (w))
972 casewriter_force_error (writer);
975 /* Returns true if an I/O error has occurred on WRITER, false otherwise. */
977 write_error (const struct sfm_writer *writer)
979 return ferror (writer->file);
982 /* Closes a system file after we're done with it.
983 Returns true if successful, false if an I/O error occurred. */
985 close_writer (struct sfm_writer *w)
996 if (w->opcode_cnt > 0)
997 flush_compressed (w);
1000 ok = !write_error (w);
1002 /* Seek back to the beginning and update the number of cases.
1003 This is just a courtesy to later readers, so there's no need
1004 to check return values or report errors. */
1005 if (ok && w->case_cnt <= INT32_MAX && !fseeko (w->file, 80, SEEK_SET))
1007 write_int (w, w->case_cnt);
1011 if (fclose (w->file) == EOF)
1015 msg (ME, _("An I/O error occurred writing system file `%s'."),
1016 fh_get_file_name (w->fh));
1018 if (ok ? !replace_file_commit (w->rf) : !replace_file_abort (w->rf))
1022 fh_unlock (w->lock);
1031 /* System file writer casewriter class. */
1032 static const struct casewriter_class sys_file_casewriter_class =
1034 sys_file_casewriter_write,
1035 sys_file_casewriter_destroy,
1039 /* Writes case C to system file W, without compressing it. */
1041 write_case_uncompressed (struct sfm_writer *w, const struct ccase *c)
1045 for (i = 0; i < w->sfm_var_cnt; i++)
1047 struct sfm_var *v = &w->sfm_vars[i];
1049 if (v->var_width == 0)
1050 write_float (w, case_num_idx (c, v->case_index));
1053 write_bytes (w, case_str_idx (c, v->case_index) + v->offset,
1055 write_spaces (w, v->padding);
1060 /* Writes case C to system file W, with compression. */
1062 write_case_compressed (struct sfm_writer *w, const struct ccase *c)
1066 for (i = 0; i < w->sfm_var_cnt; i++)
1068 struct sfm_var *v = &w->sfm_vars[i];
1070 if (v->var_width == 0)
1072 double d = case_num_idx (c, v->case_index);
1074 put_cmp_opcode (w, 255);
1075 else if (d >= 1 - COMPRESSION_BIAS
1076 && d <= 251 - COMPRESSION_BIAS
1078 put_cmp_opcode (w, (int) d + COMPRESSION_BIAS);
1081 put_cmp_opcode (w, 253);
1082 put_cmp_number (w, d);
1087 int offset = v->offset;
1090 /* This code properly deals with a width that is not a
1091 multiple of 8, by ensuring that the final partial
1092 oct (8 byte unit) is treated as padded with spaces
1094 for (width = v->segment_width; width > 0; width -= 8, offset += 8)
1096 const void *data = case_str_idx (c, v->case_index) + offset;
1097 int chunk_size = MIN (width, 8);
1098 if (!memcmp (data, " ", chunk_size))
1099 put_cmp_opcode (w, 254);
1102 put_cmp_opcode (w, 253);
1103 put_cmp_string (w, data, chunk_size);
1107 /* This code deals properly with padding that is not a
1108 multiple of 8 bytes, by discarding the remainder,
1109 which was already effectively padded with spaces in
1110 the previous loop. (Note that v->width + v->padding
1111 is always a multiple of 8.) */
1112 for (padding = v->padding / 8; padding > 0; padding--)
1113 put_cmp_opcode (w, 254);
1118 /* Flushes buffered compressed opcodes and data to W.
1119 The compression buffer must not be empty. */
1121 flush_compressed (struct sfm_writer *w)
1123 assert (w->opcode_cnt > 0 && w->opcode_cnt <= 8);
1125 write_bytes (w, w->opcodes, w->opcode_cnt);
1126 write_zeros (w, 8 - w->opcode_cnt);
1128 write_bytes (w, w->data, w->data_cnt * sizeof *w->data);
1130 w->opcode_cnt = w->data_cnt = 0;
1133 /* Appends OPCODE to the buffered set of compression opcodes in
1134 W. Flushes the compression buffer beforehand if necessary. */
1136 put_cmp_opcode (struct sfm_writer *w, uint8_t opcode)
1138 if (w->opcode_cnt >= 8)
1139 flush_compressed (w);
1141 w->opcodes[w->opcode_cnt++] = opcode;
1144 /* Appends NUMBER to the buffered compression data in W. The
1145 buffer must not be full; the way to assure that is to call
1146 this function only just after a call to put_cmp_opcode, which
1147 will flush the buffer as necessary. */
1149 put_cmp_number (struct sfm_writer *w, double number)
1151 assert (w->opcode_cnt > 0);
1152 assert (w->data_cnt < 8);
1154 convert_double_to_output_format (number, w->data[w->data_cnt++]);
1157 /* Appends SIZE bytes of DATA to the buffered compression data in
1158 W, followed by enough spaces to pad the output data to exactly
1159 8 bytes (thus, SIZE must be no greater than 8). The buffer
1160 must not be full; the way to assure that is to call this
1161 function only just after a call to put_cmp_opcode, which will
1162 flush the buffer as necessary. */
1164 put_cmp_string (struct sfm_writer *w, const void *data, size_t size)
1166 assert (w->opcode_cnt > 0);
1167 assert (w->data_cnt < 8);
1170 memset (w->data[w->data_cnt], ' ', 8);
1171 memcpy (w->data[w->data_cnt], data, size);
1175 /* Writes 32-bit integer X to the output file for writer W. */
1177 write_int (struct sfm_writer *w, int32_t x)
1179 write_bytes (w, &x, sizeof x);
1182 /* Converts NATIVE to the 64-bit format used in output files in
1185 convert_double_to_output_format (double native, uint8_t output[8])
1187 /* If "double" is not a 64-bit type, then convert it to a
1188 64-bit type. Otherwise just copy it. */
1189 if (FLOAT_NATIVE_DOUBLE != FLOAT_NATIVE_64_BIT)
1190 float_convert (FLOAT_NATIVE_DOUBLE, &native, FLOAT_NATIVE_64_BIT, output);
1192 memcpy (output, &native, sizeof native);
1195 /* Writes floating-point number X to the output file for writer
1198 write_float (struct sfm_writer *w, double x)
1201 convert_double_to_output_format (x, output);
1202 write_bytes (w, output, sizeof output);
1205 /* Writes contents of VALUE with the given WIDTH to W, padding
1206 with zeros to a multiple of 8 bytes.
1207 To avoid a branch, and because we don't actually need to
1208 support it, WIDTH must be no bigger than 8. */
1210 write_value (struct sfm_writer *w, const union value *value, int width)
1212 assert (width <= 8);
1214 write_float (w, value->f);
1217 write_bytes (w, value_str (value, width), width);
1218 write_zeros (w, 8 - width);
1222 /* Writes null-terminated STRING in a field of the given WIDTH to W. If STRING
1223 is longer than WIDTH, it is truncated; if STRING is shorter than WIDTH, it
1224 is padded on the right with spaces. */
1226 write_string (struct sfm_writer *w, const char *string, size_t width)
1228 size_t data_bytes = MIN (strlen (string), width);
1229 size_t pad_bytes = width - data_bytes;
1230 write_bytes (w, string, data_bytes);
1231 while (pad_bytes-- > 0)
1232 putc (' ', w->file);
1235 /* Recodes null-terminated UTF-8 encoded STRING into ENCODING, and writes the
1236 recoded version in a field of the given WIDTH to W. The string is truncated
1237 or padded on the right with spaces to exactly WIDTH bytes. */
1239 write_utf8_string (struct sfm_writer *w, const char *encoding,
1240 const char *string, size_t width)
1242 char *s = recode_string (encoding, "UTF-8", string, -1);
1243 write_string (w, s, width);
1247 /* Writes a record with type 7, subtype SUBTYPE that contains CONTENT recoded
1248 from UTF-8 encoded into ENCODING. */
1250 write_utf8_record (struct sfm_writer *w, const char *encoding,
1251 const struct string *content, int subtype)
1255 s = recode_substring_pool (encoding, "UTF-8", ds_ss (content), NULL);
1256 write_string_record (w, s, subtype);
1260 /* Writes a record with type 7, subtype SUBTYPE that contains the string
1263 write_string_record (struct sfm_writer *w,
1264 const struct substring content, int subtype)
1267 write_int (w, subtype);
1269 write_int (w, ss_length (content));
1270 write_bytes (w, ss_data (content), ss_length (content));
1273 /* Writes SIZE bytes of DATA to W's output file. */
1275 write_bytes (struct sfm_writer *w, const void *data, size_t size)
1277 fwrite (data, 1, size, w->file);
1280 /* Writes N zeros to W's output file. */
1282 write_zeros (struct sfm_writer *w, size_t n)
1288 /* Writes N spaces to W's output file. */
1290 write_spaces (struct sfm_writer *w, size_t n)
1293 putc (' ', w->file);